Will computers ever solve chess?

Well, admittedly, it would be pretty difficult for me to name anything that could be deemed implausible (never mind impossible) within a thousand-year time frame, so that's fair enough, I guess.

Re: 1000 years...I'm not too hopeful of mankind making it past 500 years from now given advances in bio/genetics and the inevitability of nuclear proliferation (never mind global warming/pollution).  It is simply a matter of time before a fanatic (politcal or religious) gets their finger on a button they should not.

Ergo, it's of paramount importance that humanity establish a self-sustaining colony off-world ASAP .   Terra Firma has been a single egg basket for a dangerously long time at this point.

I think that anything and everything labelled "Trump" is a complete farce...what else would anyone reasonable think?

Meh, I think it will be hard to kill all humans. There are so many and in so many places after all.

Killing the great majority, or even 99%, sure, but all humans? Someday, but not in 500 years I think.

Like most things regarding Trump, it's worthy of both a good laugh, and a good cry.

This is the best way to solve chess.

Good to hear from you @cobra91!  Very good analysis. Obviously it can only be proven that no perfect game with a 16-move checkmate exists if in fact there is none. So I would say this can be "answered" rather than "proven". Other than that I completely agree with your assessment.

hi guys me amatur

hi guys me amatur

hi guys me amatur

I'm not sure. As for the number of positions after 16 moves, and using the assumptions that Shannon used to calculate the total number of chess games, I think there would be about 10^48 positions (1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000). To the best of my knowledge nobody has cataloged them all to see if there are any forced wins by either side.

Let's put a bunch of solar panels in every desert of this planet. Then use that energy to power millions of computers to brute force their way to the solution. Saving the world and solving chess, seems like good deal, right?

That's a horrible analogy, because you are saying the goal is to travel from one known location to another (so the destination and therefore a route to get there is already known)...which is pretty much the opposite of the situation at hand.  A better analogy would be Magellan trying to circumnavigate the globe...a general direction towards a destination is posited, but nobody knows if it's even truly possible yet, and all types of obstacles will be encountered, must be charted, and avoided while still holding to the goal. 

I'm not saying how to approach the problem. I'm just mentioning the number of games. A lot of them would "junk" games, so the problem may not be as difficult as it seems. But still not easy.

To say "arguably declining" is just a horrible misrepresentation of the facts. Up until TCEC Season 9 in 2016, draw rates in the Superfinal had been steadily increasing (just as they are in World Correspondence Chess Championships), with the most recent sets of openings (that is, for seasons 7 and 8) selected by IM Erik Kislik. Then, for the Season 9 Superfinal, the selection of openings was done by Jeroen Noomen for the first time, resulting in an entirely new range of much sharper starting positions that were clearly handpicked with obvious intent. Coincidentally, the draw rate in the Season 9 Superfinal was much lower than in Season 8, and Noomen has continued to select the Superfinal openings for Seasons 10 and 11 (draw rates were slightly higher than in Season 9).

This is without even mentioning the fact that draw rates in engine matches are heavily influenced by variables such as contempt, dynamism, etc, and thus are not nearly as reliable as WCCC draw rates.

This remains to be seen, especially if such predictions are solely based on the results of the closed-door match between AlphaZero and Stockfish. It is by now quite widely known that the conditions of the match favored AlphaZero to such an extent as to reduce the event to little more than a publicity stunt. Here is a quote from Tord Romstad which accurately summarizes many of the advantages enjoyed by AlphaZero throughout those famous 100 games:

"The match results by themselves are not particularly meaningful because of the rather strange choice of time controls and Stockfish parameter settings: The games were played at a fixed time of 1 minute/move, which means that Stockfish has no use of its time management heuristics (lot of effort has been put into making Stockfish identify critical points in the game and decide when to spend some extra time on a move; at a fixed time per move, the strength will suffer significantly). The version of Stockfish used is one year old, was playing with far more search threads than has ever received any significant amount of testing, and had way too small hash tables for the number of threads."

That would probably be true from the perspective of casual fans, but complaints from experts and/or professionals in the sport would be inevitable. And for good reason -- if, for instance, the cars in such races could be operated (by their human drivers) via some form of remote control, a lot more people would be able to compete in the "sport" (could hardly be considered a sport in that case, obviously) at a very high level. The ability to both ride and operate those vehicles while moving at such high speed within a complex environment is what separates pros from amateurs when it comes to driving. A self-driving car merely emulates a human-driven car; it does not truly replicate, much less surpass, the remarkable skill of human drivers.

In any case, my point was that such an achievement would not demonstrate indisputable superiority of AI within the domain of racecar driving, as you had suggested.

I'm afraid this estimate (~10^48 possible chess positions after 16 moves by each side) is provably inaccurate. For one thing, you made no attempt to account for the astronomical number of transpositions that are possible within the scope of all legally played 16-move games. In other words, many positions are provably reachable via trillions of different 16-move games. So by counting the number of possible games with 16 full moves, you're actually counting most of the possible positions many thousands of times. Thus, an estimate of 10^48 would clearly be off the mark by at least several orders of magnitude, even though the true value is difficult to approximate without doing a lot of tedious calculations and Haskell coding.

You could have also avoided this mistake by evaluating whether or not such an estimate seems reasonable. A commonly cited upper bound for the total number of legal chess positions is 2^155 ~= 10^46.7 (shown at: https://tromp.github.io/chess/chess.html -- you'll need to scroll down to find it), and the positions that can be reached after just 16 moves constitute a mere subset of the full state space of the game. The natural conclusion is that 10^48 must be a severe overestimate of the correct value.

Something has to be said for the player aswell, the most important entity in chess. Sometimes a player will know another player so well, that he knows ways to defeat the other player that are more effective, but not necessarily correct. A very blunt example could be, how some people defeat others consistently with the socalled scholar's mate. Now, if they know that the opponent doesn't know how to defend against it, they can mate him in 4 moves. But a computer would never attempt this, because it always assumes the opponent is playing the top notch best moves and so it tries to maximize it's game, but at the same time, not knowing that this particular player always falls for the scholar's mate, the computer even top notch is not as effective as the almost beginner player who knows that his other very much so beginner player always falls for mate in 4 moves. Now, what am I trying to say here, merely that even though the game of chess get's solved, there is something to be said for knowing your enemy, which according to some is the first rule of war. Some players are very good at prodding out the weak points in other players play, and so will against said players be most effective whereas a computer that always go for the objectively best moves will not. This I believe is a form of anticipation and psychology in chess, a form of sense of prophylaxis. Tal was also a good example of this, since it is said that the positions might seem quiet and dull, and suddenly there is a flurry of moves, and some stuff is sacrificed and then he had won. Objectively a computer might say the moves were suboptimal, but they still were highly effective against human opponents. So even if chess gets solved to a win, loss or draw, chess will not die out as long as there are people playing people.

The number of games though is way more, like 10^1000000000000 prolly. And without the 50 move rule, exponentially more! Even quantum computers might have trouble.

You're right, I calculated games and not positions. Thanks for pointing that out. I used the same assumptions that Shannon used to calculate the total number of games. See paper below.

(Shannon paper)
While transiting the game tree to 16 moves, some checkmates will also be reached prior to the 16th move. It would also be interesting to see if any of those games (shorter than 16 moves) are perfect games too.

" The seeds can be pressed for their oil, then planted to grow many many many more plants."

So Scott, are you saying that pressed seeds can still grow into viable plants? 

Now, I accept that seeds from grapes which have been pressed will still be viable (as long as the grapes were cold pressed)